http://bugzilla.spamassassin.org/show_bug.cgi?id=3225
------- Additional Comments From [EMAIL PROTECTED] 2004-04-23 20:10 ------- Subject: Re: RFE: Bayes optimizations On Fri, Apr 23, 2004 at 04:59:41PM -0700, [EMAIL PROTECTED] wrote: > resource usage. Much of that was from updating the atime fields of every token > that was selected to be used in a Bayes calculation of a message. I can buy that to some degree. It's the same reason you turn off atime updating on filesystems for things like proxy server and news server storage. However, as with news and proxy servers, they'll typically do expiry with data from the files themselves -- which for us _is_ the atime update. Do we have a real breakdown as to where our main I/O usage comes from? > I will concede that the bigger factor there is update frequency, which has to > do > with the granularity of atime, not its size. So we could stick to 32 bit > standard unix time format and be safe until 2038. If we could have a > configurable scaling factor for atime then it would control the granularity > very > simply. Well, we already have that with the current code. atime updates go into the journal which is "light" I/O usage, then periodically the journal gets synced to the DB. The code that reads in the journal caches atime updates, and then does 1 update per token instead of multiple updates per token. ------- You are receiving this mail because: ------- You are the assignee for the bug, or are watching the assignee.
